A Brief Observation to Screen Autism in Toddlers and Predict Developmental Trajectory.

Objective: Autism spectrum disorder (ASD) affects approximately 1 in 31 children. Early diagnosis is crucial for optimizing outcomes through early interventions, and primary care settings need efficient tools to identify children presenting autistic features. This study explores the potential of early socio-communicative behaviors, measured by the Early Social Communication Scales, to screen for ASD in children younger than 3 years old, and predict their future cognitive development using machine learning models. Study design: This study analyzed longitudinal data from 113 children with ASD and 59 with typical development (TD), aged from 1 to 3 at baseline. Twenty-three ESCS variables were used to screen for ASD and predict cognitive development. The C5.0 decision tree algorithm was used to classify ASD vs TD, while linear regression and K-means clustering identified cognitive development patterns among autistic children. K-fold cross-validation, permutation testing, and undersampling were used for validation. Results: We distinguished between ASD and TD children with 95% accuracy, 96% sensitivity and 92% specificity. Nine behaviors contributed to distinguish ASD from TD. Behaviors that contributed most are the child's ability to initiate a turn taking and to point at desired objects. A separate model stratified children into groups with different cognitive outcome with 97% accuracy. Behavioral requests variables contributed in distinguishing extreme cognitive trajectories in autistic children. Conclusions: We provide an original decision-algorithm focusing on early socio-communicative behaviors to guide pediatricians through autism screening and cognitive development prediction.